The feasibility of deep learning-based synthetic contrast-enhanced CT from nonenhanced CT in emergency department patients with acute abdominal pain

Se Woo Kim, Jung Hoon Kim, Suha Kwak, Minkyo Seo, Changhyun Ryoo, Cheong‐Il Shin, Siwon Jang, Jungheum Cho, Young-Hoon Kim, Kyutae JeonScientific 2021 · DOI

Abstract

This study investigates the feasibility of deep learning-based synthetic contrast-enhanced CT (DL-SCE-CT) generated from non-enhanced CT (NECT) for patients presenting to the emergency department (ED) with acute abdominal pain (AAP). Using a two-stage generative adversarial network, synthetic images were produced for 353 patients. Six radiologists (three experienced, three in training) evaluated cases using NECT alone and then with DL-SCE-CT. Results showed significant improvements in diagnostic accuracy for 50% of reviewers, particularly trainees, and a significant boost in confidence for 83.3% of reviewers across diagnosis and disposition decisions.

Introduction

Acute abdominal pain (AAP) is a frequent cause for emergency department (ED) visits. While contrast-enhanced CT (CECT) is the diagnostic gold standard, the use of intravenous contrast agents is contraindicated in patients with risks such as chronic kidney disease or severe allergies. Non-enhanced CT (NECT) is a safer alternative but often lacks the diagnostic clarity required for definitive decision-making. This research explores whether deep learning can bridge this gap by generating synthetic contrast-enhanced images (DL-SCE-CT) from NECT data.

Methods

The primary technical challenge in training synthetic contrast models is image misalignment due to patient motion between scans. To address this, the researchers employed a two-stage training strategy using a Conditional Generative Adversarial Network (cGAN) with SPADE architecture:

  1. Inverse Mapping: A generator was trained to create synthetic NECT from real CECT. This ensured perfect alignment between the synthetic input and the ground truth contrast image.
  2. Target Mapping: A second generator was trained to transform the synthetic NECT back into a synthetic CECT.

The model was trained on 35,414 paired images and validated on a multi-institutional dataset of 353 patients. Six radiologists (ER1-3: experienced; TR1-3: trainees) performed evaluations in two sessions: first using NECT alone, then using NECT combined with DL-SCE-CT.

Key Results

The integration of DL-SCE-CT led to measurable improvements in both performance and subjective certainty. Diagnostic accuracy improved significantly for three out of six radiologists (p < 0.05), with the most pronounced gains observed among the training radiologists (TR2 and TR3).

Comparison of diagnostic and disposition accuracy and confidence across two sessions (Session 1: NECT alone; Session 2: NECT + DL-SCE-CT).

Confidence levels saw the most consistent improvement. Five out of six radiologists reported significantly higher confidence in both their diagnoses and their disposition decisions (p < 0.001). Subgroup analysis revealed that the AI was particularly helpful in cases involving underlying malignancies and miscellaneous medical conditions, where NECT typically struggles to provide clear visualization.

Discussion

The study demonstrates that DL-SCE-CT can serve as a "trainee force multiplier," helping less experienced clinicians achieve diagnostic performance closer to that of senior radiologists. The two-stage training approach effectively bypassed the need for complex deformable registration, providing a robust method for generating synthetic medical imagery.

However, the authors note the risk of "automation bias," where increased confidence might occur even in incorrect diagnoses. Future work should focus on integrating these synthetic images into clinical workflows and assessing their impact on long-term patient outcomes, such as mortality and length of stay.

Conclusion

DL-SCE-CT is a feasible and effective tool for enhancing the diagnostic process in the ED for patients with acute abdominal pain. By improving both accuracy and confidence—especially for trainees and in complex subgroups—this technology offers a promising, contrast-free alternative for high-risk patients.

Summary updated

Abstract

Our objective was to investigate the feasibility of deep learning-based synthetic contrast-enhanced CT (DL-SCE-CT) from nonenhanced CT (NECT) in patients who visited the emergency department (ED) with acute abdominal pain (AAP). We trained an algorithm generating DL-SCE-CT using NECT with paired precontrast/postcontrast images. For clinical application, 353 patients from three institutions who visited the ED with AAP were included. Six reviewers (experienced radiologists, ER1-3; training radiologists, TR1-3) made diagnostic and disposition decisions using NECT alone and then with NECT and DL-SCE-CT together. The radiologists' confidence in decisions was graded using a 5-point scale. The diagnostic accuracy using DL-SCE-CT improved in three radiologists (50%, P = 0.023, 0.012, < 0.001, especially in 2/3 of TRs). The confidence of diagnosis and disposition improved significantly in five radiologists (83.3%, P < 0.001). Particularly, in subgroups with underlying malignancy and miscellaneous medical conditions (MMCs) and in CT-negative cases, more radiologists reported increased confidence in diagnosis (83.3% [5/6], 100.0% [6/6], and 83.3% [5/6], respectively) and disposition (66.7% [4/6], 83.3% [5/6] and 100% [6/6], respectively). In conclusion, DL-SCE-CT enhances the accuracy and confidence of diagnosis and disposition regarding patients with AAP in the ED, especially for less experienced radiologists, in CT-negative cases, and in certain disease subgroups with underlying malignancy and MMCs.
BibTeX
@inproceedings{kim2021feasibility,
  author = {Se Woo Kim and Jung Hoon Kim and Suha Kwak and Minkyo Seo and Changhyun Ryoo and Cheong‐Il Shin and Siwon Jang and Jungheum Cho and Young-Hoon Kim and Kyutae Jeon},
  title = {{The feasibility of deep learning-based synthetic contrast-enhanced CT from nonenhanced CT in emergency department patients with acute abdominal pain}},
  booktitle = {Scientific},
  year = {2021},
  doi = {10.1038/s41598-021-99896-4},
}
Figure